Using ARM on Coiled#

Coiled supports ARM (Graviton) instance types on AWS.

We found that for some workloads, ARM is significantly faster and cheaper. We don’t yet know how to predict whether a specific workload will be better on ARM, but we encourage you to try for yourself!


Coiled offers a variety of ways to get the Python packages you need installed on your cluster:

  1. You can use Package Synchronization to have us install packages based on what you have in your local environment.

  2. You can use coiled.create_software_environment() and specify a list of packages to install with pip or conda.

  3. You can use a pre-built Docker image.

Currently only (3) is supported for ARM instances—support for other options is on our TODO list.

You can either build your own Docker image (harder) or you can use the coiled-runtime 0.1.1 image which is already built for both x86 and ARM.

Starting a cluster#

If coiled-runtime has the packages you need, then it’s easiest to use that on the cluster. You can first install it locally, using mamba to create your Python environment:

mamba create -n runtime-021 python=3.10 coiled-runtime=0.2.1
mamba activate runtime-021

Next, you can create a Coiled Python environment using the coiled-runtime image, publicly available on DockerHub.

import coiled

# create a Coiled environment
    name="coiled-runtime-0-2-1-py310", container="coiled/coiled-runtime:0.2.1-py3.10"

Then, you can create a cluster from wherever you run Python:

cluster = coiled.Cluster(
    # ARM instance types for scheduler and workers
    # specify the environment you created above

You can select instance types based on the recommendations above, and request as many workers as you want!

If coiled-runtime doesn’t have what you need, build your multi-arch Docker image (see Using a custom Docker image below), create a Coiled software environment from that image, and then use that on your cluster (see Docker images).

Using a custom Docker image#

You can use GitHub Actions to build your own multi-arch (x86 and ARM) Docker image (see this example yaml file).

Locally, you can build multi-arch images like this:

docker buildx build --platform linux/arm64,linux/amd64 \
-t <your-image-name>:latest --push .

You can then upload your local image to wherever you usually store Docker images (e.g. Docker Hub or Amazon ECR) and create a Coiled Python environment using this image (see Docker images for instructions).

Instance types#

For common Intel instance types you might be using already, there’s a roughly equivalent ARM instance type you could try instead.






burstable (best for interactive/non-compute-intensive work)



non-burstable balanced compute/memory (sensible default for common workloads)



compute-optimized (higher ratio of vCPUs to memory)



memory-optimized (higher ratio of memory to vCPUs)

The *7g families have Graviton 3 and are slightly more expensive than corresponding *6g family with older generation ARM processor, though Graviton 3 is significantly better and is still cheaper (typically by 15%) than corresponding Intel instance type.


Only the c7g family has the newest generation Graviton3 processors, and these are significantly better than previous generations. But c7g is “compute optimized”, so it has less memory per instance than equivalent m6i. Using next size up with c7g means you’ll have as much memory per instance. Cost per instance/hour is higher, but if your workload completes in significantly less time—which we’ve seen for some workloads—your total cost will still be lower.